File size: 5,209 Bytes
42b2b3c a5c9fa3 42b2b3c a5c9fa3 42b2b3c a5c9fa3 42b2b3c a5c9fa3 42b2b3c a5c9fa3 42b2b3c a5c9fa3 42b2b3c a5c9fa3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 | import streamlit as st
from search.search_classical import classical_search, classical_retrieve_chunks
from search.search_best_pair import best_pair_search
from llm import generate_response
def pick_mode(label: str) -> str:
if label.startswith("Semantic"):
return "semantic"
if label.startswith("Keyword"):
return "bm25"
return "hybrid"
st.set_page_config(page_title="🔍 Multimodal Search The Batch")
st.image("data/the-batch-logo.webp", width=300)
st.title("Multimodal Assistant")
mode = st.selectbox("🔎 Select the search mode:", ["Classical RAG", "Multimodal RAG"])
query = st.text_input("📝 Enter the text query:")
# --- Classical controls ---
classical_retriever = "Semantic (embeddings)"
use_reranker = True
if mode == "Classical RAG":
classical_retriever = st.radio(
"🧩 Classical retrieval:",
["Semantic (embeddings)", "Keyword (BM25)", "Hybrid (BM25 + Semantic)"],
horizontal=True
)
use_reranker = st.checkbox("✨ Use reranker (cross-encoder)", value=True)
# --- Preview results ---
results = []
if query:
if mode == "Classical RAG":
search_mode = pick_mode(classical_retriever)
results = classical_search(query, k=3, mode=search_mode)
else:
results = best_pair_search(query, k=3)
st.markdown(f"### 📄 Results found: {len(results)}")
for i, meta in enumerate(results):
st.markdown(f"### 🔹 Result {i + 1}")
if meta.get("title"):
st.markdown(f"**📖 Name:** {meta['title']}")
if meta.get("date"):
st.markdown(f"**📅 Date of publication:** {meta['date']}")
if meta.get("description"):
st.markdown(f"**📝 Description:** {meta['description']}")
if meta.get("image_url"):
st.image(meta["image_url"], use_container_width=True)
if meta.get("content"):
st.markdown("**📚 Part of the article:**")
st.write(meta["content"][:500] + "...")
if meta.get("source_url"):
st.markdown(f"[🔗 Read the full article →]({meta['source_url']})")
st.markdown("---")
# --- Generate answer ---
if query and st.button("🧠 Generate a response to a query"):
if mode == "Classical RAG":
search_mode = pick_mode(classical_retriever)
chunks = classical_retrieve_chunks(
query=query,
mode=search_mode,
fetch_k=50,
rerank_k=5,
use_reranker=use_reranker
)
docs = []
for idx, c in enumerate(chunks, start=1):
meta = c.get("metadata", {})
docs.append({
"id": idx,
"title": meta.get("title", ""),
"description": meta.get("description", ""),
"source_url": meta.get("source_url", ""),
"content": c.get("chunk_text", ""),
"retriever": c.get("retriever", ""),
"rerank_score": c.get("rerank_score", None),
})
response = generate_response(query, docs)
st.markdown("### 🤖 Generated Response:")
st.success(response)
st.markdown("### 📌 Sources")
for d in docs:
st.markdown(f"**[{d['id']}] {d.get('title','')}**")
if d.get("source_url"):
st.markdown(d["source_url"])
st.write((d.get("content") or "")[:450] + "...")
if d.get("retriever"):
st.caption(f"retriever: {d['retriever']}")
if d.get("rerank_score") is not None:
st.caption(f"rerank_score: {d['rerank_score']:.4f}")
st.markdown("---")
else:
# ✅ Multimodal mode:
# Preview stays multimodal (best_pair_search),
# but the ANSWER is generated from TEXT chunks (hybrid) for reliable QA + citations.
chunks = classical_retrieve_chunks(
query=query,
mode="hybrid",
fetch_k=50,
rerank_k=5,
use_reranker=True
)
docs = []
for idx, c in enumerate(chunks, start=1):
meta = c.get("metadata", {})
docs.append({
"id": idx,
"title": meta.get("title", ""),
"description": meta.get("description", ""),
"source_url": meta.get("source_url", ""),
"content": c.get("chunk_text", ""),
"retriever": c.get("retriever", ""),
"rerank_score": c.get("rerank_score", None),
})
response = generate_response(query, docs)
st.markdown("### 🤖 Generated Response:")
st.success(response)
st.markdown("### 📌 Sources (text chunks)")
for d in docs:
st.markdown(f"**[{d['id']}] {d.get('title','')}**")
if d.get("source_url"):
st.markdown(d["source_url"])
st.write((d.get("content") or "")[:450] + "...")
if d.get("retriever"):
st.caption(f"retriever: {d['retriever']}")
if d.get("rerank_score") is not None:
st.caption(f"rerank_score: {d['rerank_score']:.4f}")
st.markdown("---")
|